Search NASA⌕ Search

SEARCH · Search NASA

Results for “deep data”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 199 records · Page 11

Deep Lynx MOOSE Adapter

The Deep Lynx MOOSE Adapter connects the Deep Lynx data warehouse with any MOOSE executable. The Adapter can receive events from Deep Lynx and will take incoming data to format a template input file for the MOOSE executable. Returns from the MOOSE run are sent back to Deep Lynx for use by other applications.

Browning, JerenM↗

Probing the transition from shallow to deep convection using ASR data and large-eddy simulations

Over the duration of this project, we have made the following advances. 1) We have developed a novel approach to obtain a Lagrangian view of convection from high-resolution numerical model through Lagrangian tracking. This approach nicely complements the more traditionally used Eulerian statistics. We have applied this approach to a range of problem. 2) We have looked into improving and extending our parameterizations based on stochastically entraining parcels, developed previously for shallow convection. 3) This grant also supported our effort on a paper where we compared cumulus parameterizations and cloud resolving models in terms of their linear response functions. This work will help the community to better evaluate and develop cumulus parameterization. 4) We have applied Lagrangian tracking to shallow convection, deep convection with and without convective organization to better characterize their dynamics and the transition between them. 5) We have devised a novel way of using Lagrangian to identify cold pools, an area identified as of great interest by the ASR community. Our algorithm has a number of advantages and in particular can handle merging cold pools more gracefully than existing techniques. 6) We demonstrated that we can, for the first time, correctly reproduce both the diurnal and seasonal cycle of the hydrologic cycle in the Amazon using a strategy that explicitly represents convection but parameterizes large-scale circulation. In addition we showed that the main cause of the wet season is the presence of an early morning fog, which insulate the surface from top of the atmosphere shortwave radiation. In essence this fog makes the day shorter because radiation cannot penetrate to the surface in the early morning. This is why all fluxes are reduced in the wet season compared to the dry season. 7) We have investigated the life cycle of cold pools and the role of surface diabatic heating. We show that surface heating can kill cold pols and reduce the number of large cold pools and the organization of convection. The effect is quite dramatic over land where the entire distribution of cold pools is modified, and the cold pools are much warmer and more humid with surface diabatic heating below the cold pools. The PI and the co-PI continue to work together on parameterization of cold pools.

54 ENVIRONMENTAL SCIENCES↗

Domain Adaptation for Measurements of Strong Gravitational Lenses

Upcoming surveys are predicted to discover galaxy-scale strong lenses on the magnitude of 105, making deep learning methods necessary in lensing data analysis. Currently, there is insufficient real lensing data to train deep learning algorithms, but training only on simulated data results in poor performance on real data. Domain adaptation can bridge the gap between simulated and real datasets. We adopt domain adaptation on the estimation of Einstein radius in simulated galaxy-scale gravitational lensing images. We evaluate two domain adaptation techniques - domain adversarial neural networks (DANN) and maximum mean discrepancy (MMD). We train on a source domain of simulated lenses and apply it to a target domain with emulation of DES survey conditions. We show that both domain adaptation techniques can significantly improve the model performance on the more complex target domain datasets. Our results show the potential of using domain adaptation to perform analysis on future survey data with a deep neural network trained on simulated data.

79 ASTRONOMY AND ASTROPHYSICS↗

Quantifying the generalization error in deep learning in terms of data distribution and neural network smoothness

We report the accuracy of deep learning, i.e., deep neural networks, can be characterized by dividing the total error into three main types: approximation error, optimization error, and generalization error. Whereas there are some satisfactory answers to the problems of approximation and optimization, much less is known about the theory of generalization. Most existing theoretical works for generalization fail to explain the performance of neural networks in practice. To derive a meaningful bound, we study the generalization error of neural networks for classification problems in terms of data distribution and neural network smoothness. We introduce the cover complexity (CC) to measure the difficulty of learning a data set and the inverse of the modulus of continuity to quantify neural network smoothness. A quantitative bound for expected accuracy/error is derived by considering both the CC and neural network smoothness. Although most of the analysis is general and not specific to neural networks, we validate our theoretical assumptions and results numerically for neural networks by several data sets of images. The numerical results confirm that the expected error of trained networks scaled with the square root of the number of classes has a linear relationship with respect to the CC. We also observe a clear consistency between test loss and neural network smoothness during the training process. In addition, we demonstrate empirically that the neural network smoothness decreases when the network size increases whereas the smoothness is insensitive to training dataset size.

97 MATHEMATICS AND COMPUTING↗

AtomSets as a hierarchical transfer learning framework for small and large materials datasets

Abstract Predicting properties from a material’s composition or structure is of great interest for materials design. Deep learning has recently garnered considerable interest in materials predictive tasks with low model errors when dealing with large materials data. However, deep learning models suffer in the small data regime that is common in materials science. Here we develop the AtomSets framework, which utilizes universal compositional and structural descriptors extracted from pre-trained graph network deep learning models with standard multi-layer perceptrons to achieve consistently high model accuracy for both small compositional data (<400) and large structural data (>130,000). The AtomSets models show lower errors than the graph network models at small data limits and other non-deep-learning models at large data limits. They also transfer better in a simulated materials discovery process where the targeted materials have property values out of the training data limits. The models require minimal domain knowledge inputs and are free from feature engineering. The presented AtomSets model framework can potentially accelerate machine learning-assisted materials design and discovery with less data restriction.

Chen, Chi (ORCID:0000000180087043)↗

Deep Borehole Disposal Research: Geological Data Evaluation Alternative Waste Forms and Borehole Seals

This report documents deep borehole disposal research during FY2014, as directed by U.S. Department of Energy (DOE) Used Fuel Disposition (UFD) Campaign. These research efforts are principally directed at advancing the deep borehole disposal project to the implementation of a full-scale Research, Development, and Demonstration (RD&D) project. Activities of particular relevance to this goal include evaluation of guidelines for selection of a deep borehole field test site, analyses of deep borehole disposal of alternative waste forms, and technical planning for borehole seals research.

Arnold, Bill W.↗

Deep glassy state dynamic data challenge glass models: Configurational entropy models

This work was the result of the student's (Dongjie Chen) qualifying exam response. In it the experimental observation from our labs that the relaxation times of ultrastable glasses (a 20 million year old amber and a vapor deposited amorphous teflon) do not follow the expected Vogel-Fulcher or WLF types of divergence as the temperatures fall below the glass transition temperature. In the work we evaluated several theories by considering them in different ways. We used them as generally used and show that they do not capture the non-diverging behavior. Because the models were all entropy-based models, we also considered the possibility that the temperature dependence of the entropy used in the models is incorrect and chose two ways of estimating the entropy that do not follow the classical theoretical models. First we used the actually measured calorimetric response for the two ultrastable glasses, and we used the Milchev model, which does not go to zero above 0 Kelvin. Both of these approaches led to improved agreement between the theories and the experimental data. Finally, we also asked what the temperature dependence of the entropy would have to be so that the models agreed with the observed data. In some cases the entropy so-calculated was reasonable, while in others it was unphysical. In sum, the important point of the work is that the observation of non-diverging time-scales that suggest that there is no ideal glass transition can be reasonably captured by current theories if they take more realistic models of the entropy than is usually the case.

36 MATERIALS SCIENCE↗

COMPARING GC AND FIELD LMXBs IN ELLIPTICAL GALAXIES WITH DEEP CHANDRA AND HUBBLE DATA

We present a statistical study of the low-mass X-ray binary (LMXB) populations of three nearby, old elliptical galaxies: NGC 3379, NGC 4278, and NGC 4697. With a cumulative {approx}1 Ms Chandra ACIS observing time, we detect 90-170 LMXBs within the D{sub 25} ellipse of each galaxy. Cross-correlating Chandra X-ray sources and HST optical sources, we identify 75 globular cluster (GC) LMXBs and 112 field LMXBs with L{sub X} > 10{sup 36} erg s{sup -1} (detections of these populations are 90% complete down to luminosities in the range of 6 x 10{sup 36} to 1.5 x 10{sup 37} erg s{sup -1}). At the higher luminosities explored in previous studies, the statistics of this sample are consistent with the properties of GC-LMXBs reported in the literature. In the low-luminosity range allowed by our deeper data (L{sub X} < 5 x 10{sup 37} erg s{sup -1}), we find a significant relative lack of GC-LMXBs, when compared with field sources. Using the co-added sample from the three galaxies, we find that the incompleteness-corrected X-ray luminosity functions (XLFs) of GC and field LMXBs differ at {approx}4{sigma} significance at L{sub X} < 5 x 10{sup 37} erg s{sup -1}. As previously reported, these XLFs are consistent at higher luminosities. The presently available theoretical models for LMXB formation and evolution in clusters are not sophisticated enough to provide a definite explanation for the shape of the observed GC-LMXB XLF. Our observations may indicate a potential predominance of GC-LMXBs with donors evolved beyond the main sequence, when compared to current models, but their efficient formation requires relatively high initial binary fractions in clusters. The field LMXB XLF can be fitted with either a single power-law model plus a localized excess at a luminosity of (5-6) x 10{sup 37} erg s{sup -1}, or a broken power law with a similar low-luminosity break. This XLF may be explained with NS-red-giant LMXBs, contributing to {approx}15% of total LMXBs population at {approx}5 x 10{sup 37} erg s{sup -1}. The difference in the GC and field XLFs is consistent with different origins and/or evolutionary paths between the two LMXB populations, although a fraction of the field sources are likely to have originated in GCs.

79 ASTRONOMY AND ASTROPHYSICS↗

NeuralCubes: Deep Representations for Visual Data Exploration

Visual exploration of large multi-dimensional datasets has seen tremendous progress in recent years, allowing users to express rich data queries that produce informative visual summaries, all in real time. Techniques based on data cubes are some of the most promising approaches. However, these techniques usually require a large memory footprint for large datasets. To tackle this problem, we present NeuralCubes: neural networks that predict results for aggregate queries, similar to data cubes. NeuralCubes learns a function that takes as input a given query, for instance, a geographic region and temporal interval, and outputs the result of the query. The learned function serves as a real-time, low-memory approximator for aggregation queries. Our models are small enough to be sent to the client side (e.g. the web browser for a web-based application) for evaluation, enabling data exploration of large datasets without database/network connection. Here, we demonstrate the effectiveness of NeuralCubes through extensive experiments on a variety of datasets and discuss how NeuralCubes opens up opportunities for new types of visualization and interaction.

97 MATHEMATICS AND COMPUTING↗